The current race to build autonomous AI agents has shifted from a battle of model intelligence to a battle of connectivity. Every enterprise is currently scrambling to figure out how to let an LLM securely touch their internal databases without creating a catastrophic security hole. This tension has turned the product trial phase into a high-stakes gamble, where the line between a potential partnership and industrial espionage becomes dangerously thin. This week, that tension exploded into a legal battle that could serve as a warning for every AI infrastructure startup dealing with tech-heavy enterprise clients.

The Anatomy of a Trade Secret Dispute

Runlayer, an AI infrastructure startup that has secured 42 million dollars in funding from heavyweights like Khosla Ventures and Felicis, has filed a lawsuit against the HR software giant Rippling. The complaint alleges trade secret misappropriation, unfair competition, and breach of contract. At the heart of the conflict is the MCP Gateway, a product Runlayer launched mid-last year to streamline how AI models interact with external data.

According to the legal filings, the relationship began as a standard product trial. Rippling entered the picture as a potential customer, leading to an intensive engineering collaboration that spanned approximately one year. During this period, Runlayer claims it provided Rippling with deep access to its inner workings, including detailed product roadmaps and actual source code. This exchange was governed by a mutual non-disclosure agreement (NDA) and a trial agreement that explicitly prohibited the replication of intellectual property or the creation of derivative works.

The relationship soured when negotiations over pricing collapsed. Runlayer alleges that immediately after the trial ended, internal communications from Rippling revealed the existence of an internal project designed to clone Runlayer's product almost one-to-one. Rippling has denied these allegations, asserting that its current offerings are the result of its own proprietary information and that it has simply built a superior tool for connecting business data to AI.

The Implementation Gap in Open Protocols

To understand why this dispute is so volatile, one must look at the technical foundation of the product: the Model Context Protocol (MCP). Released by Anthropic in November 2024, MCP is an open-source protocol designed to act as a universal building block for interoperability between different AI models and tools. Because the protocol itself is open, the value in the market has shifted from the protocol to the gateway implementation.

An MCP Gateway is not merely a pass-through for the protocol; it is a sophisticated control layer. It manages agent permissions, handles complex authentication, and acts as a security filter that monitors data flow between the AI model and the enterprise's sensitive internal systems. In a corporate environment, this gateway is the only thing preventing an AI agent from accidentally leaking payroll data or deleting a database during a hallucination. It transforms a raw protocol into a production-ready enterprise tool.

This creates a specific kind of vulnerability for startups. When a company like Rippling, which possesses massive internal engineering resources, conducts a deep-dive trial, they aren't just testing the output of the software. They are analyzing the implementation logic. Because the underlying MCP protocol is public, the only remaining "secret sauce" is how the gateway manages state, security, and scale. If a client gains access to the source code under the guise of a trial, the barrier to building an in-house version drops significantly. The conflict here is a classic clash between the buy-versus-build strategy, where the client uses a trial to determine if they can build the solution themselves more cheaply than they can buy it.

This case exposes a systemic risk in the enterprise AI pipeline. When startups share source-level details to prove technical viability to a sophisticated buyer, they are essentially handing over the blueprint for their own replacement. The legal protections of an NDA are often a thin shield against a company with the engineering capacity to rewrite a feature from scratch based on a conceptual understanding of the source code.

This litigation will likely define the boundaries of intellectual property for the next generation of AI middleware providers.